Google DeepMind's Gemini Robotics 2 gives humanoid robots full-body control from feet to fingertips

{ "TITLE": "Google DeepMind's Gemini Robotics 2 gives humanoid robots whole-body control from feet to fingertips", "BODY": "Google DeepMind on Thursday announced Gemini Robotics 2, a new AI model that can control a humanoid robot's entire body from its feet to its fingertips, marking a significant leap beyond the previous version that only handled upper-body movements. The update, published on July 30, 2026, at 5:18 PM UTC, also includes improvements to the Gemini Robotics ER 2 vision-language model and the On-Device Model, which runs locally without an internet connection. The announcement comes as part of a broader push by Google DeepMind to make robots more capable in real-world environments, with demonstrations showing robots performing tasks that require precise coordination and safety awareness.\n\n## Whole-body coordination unlocks new tasks\n\nThe previous Gemini Robotics model was limited to controlling only the upper body of humanoid robots. With Gemini Robotics 2, robots can now walk, crouch, stretch, and manipulate objects using their full range of motion. Google DeepMind says the latest version can "control entire humanoid robots," enabling actions that require coordination from head to toe. In demonstrations, Apptronik's Apollo 2 robot bent over to pick up a watering can and found and took specific items off a shelf. The model also supports control of more complex, five-fingered hands, allowing robots to perform tasks like sealing a Ziploc bag, tying a trash bag, and unscrewing a lightbulb. Google DeepMind notes that robots "have more to advance in movement speed," but says this update "is an important step towards the skills needed to complete more complex, real-world tasks that require whole-body coordination." The ability to use both legs and arms simultaneously opens up new possibilities for robots to navigate cluttered spaces, such as warehouses or homes, where bending, reaching, and balancing are essential. The Apollo 2 robot, built by Apptronik, served as the primary test platform for these demonstrations, showcasing how the model can handle tasks that were previously impossible with only upper-body control.\n\n## Gemini Robotics ER 2 improves safety and collaboration\n\nThe updated Gemini Robotics ER 2 is a vision-language model designed for embodied reasoning. It helps robots analyze their surroundings, process instructions, and perform multi-step tasks over extended periods. The model now understands when tasks begin and end, and it allows multiple robots of different types to work together. One video shows Apollo 2 instructing Google's dual-arm robot to put tools inside a bin while cleaning a garage. Google DeepMind describes Gemini Robotics ER 2 as its safest robotics model to date. The model can "better detect when humans are nearby, trigger safety tool calls and bring the robot to a safe stop if someone approaches too closely." This safety feature is critical for deploying robots in environments where people work alongside them, such as factories or homes. The model also improves long-horizon task execution, meaning robots can follow complex sequences of instructions without losing track of what they are doing. For example, a robot could be told to clean a room, and it would understand the steps involved, from picking up objects to wiping surfaces, without needing constant human guidance. The collaboration between different robot types, such as the Apollo 2 and Google's dual-arm robot, demonstrates how the AI can coordinate actions across multiple machines, potentially allowing teams of robots to tackle larger projects like assembling furniture or organizing storage units.\n\n## On-Device Model adapts to new robot bodies\n\nThe Gemini Robotics On-Device Model runs locally on a robot without needing an internet connection. It has been updated to adapt faster to new embodiments, including those with "drastically different shapes, sensors and degrees of freedom." This means the same AI can quickly adjust to control robots that look and move very differently from one another. The On-Device Model is designed to be lightweight enough to run on the robot's onboard hardware, reducing latency and ensuring that the robot can operate even in areas with poor connectivity. This adaptability is key for manufacturers who want to use the same AI software across multiple robot designs, from humanoid robots like Apollo 2 to wheeled or tracked robots. Google DeepMind has not disclosed the specific hardware requirements for running the model, but the company emphasizes that it can be deployed on a range of platforms. The ability to adapt to new embodiments without extensive retraining could accelerate the adoption of AI in robotics, as companies can integrate the model into existing robot fleets without overhauling their hardware.\n\n## Broader implications for real-world deployment\n\nThe combination of whole-body control, improved safety, and faster adaptation to new hardware suggests Google DeepMind is pushing toward robots that can operate more autonomously in unstructured environments. While the company acknowledges that movement speed still needs improvement, the ability to coordinate feet, torso, arms, and fingers opens up tasks like cleaning, organizing, and handling delicate objects that were previously out of reach. The Apollo 2 robot, built by Apptronik, served as the primary test platform for the demonstrations. Google's dual-arm robot also participated in collaborative tasks, showing that different robot types can work together under the same AI system. The updates to Gemini Robotics ER 2 and the On-Device Model further enhance the system's utility in real-world settings, where robots must interact with humans and adapt to changing conditions. Google DeepMind has not announced a timeline for commercial deployment, but the demonstrations indicate that the technology is advancing rapidly. The ability to perform tasks like tying a trash bag or unscrewing a lightbulb, while seemingly simple, requires precise force control and spatial awareness, which are challenging for robots. The company's focus on safety, including automatic stops when humans approach, addresses one of the key barriers to widespread robot adoption. As robots become more capable, they could take on roles in logistics, healthcare, and home assistance, though significant work remains to improve speed and reliability.\n\n## Related on Neura Market\n\n- Google DeepMind robotics research\n- Humanoid robot industry developments\n- AI model updates and breakthroughs" }